Researchers developed a rapid test to predict green-aroma compound levels in wine grapes

Researchers reported a rapid method to predict levels of 3-isobutyl-2-methoxypyrazine, or IBMP, in Chilean red wine grapes, a compound tied to the vegetal or “green” aromas that can lower the perceived quality of some wines when present at high levels.

The study, published June 4 in OENO One, tested absorbance-transmittance excitation-emission matrix spectroscopy, known as A-TEEM, together with machine learning models as an alternative screening tool to the gas chromatography-mass spectrometry methods that are now commonly used to measure IBMP. Those standard methods are highly sensitive, but they are also costly, slow and technically demanding.

The work focused on four red grape varieties grown in Chile: Cabernet-Sauvignon, Carménère, Merlot and Cabernet franc. According to the paper, the researchers analyzed samples collected during four harvest years, from 2020 to 2022 and again in 2024, across 72 vineyards. In total, they prepared 2,400 grape extract samples for A-TEEM analysis and used reference measurements from more than 550 grape samples obtained with SPME-GC-MS/MS, a standard laboratory technique for ultratrace aroma compounds.

IBMP matters because it accumulates in immature grapes and tends to decline as fruit ripens. In red wines, especially Cabernet-Sauvignon, it is associated with green pepper and other vegetal notes. The paper notes that the sensory threshold in red wine is generally reported between 2 and 16 ng/L, which means even very small amounts can affect aroma.

That low threshold is one reason wineries and growers monitor the compound before harvest. But sending samples to specialized laboratories can take days or longer, which can limit decisions on picking dates or vineyard interventions. A faster test could help producers judge ripeness more quickly and manage a sensory marker that has direct consequences for wine style and market acceptance.

The authors said their goal was to develop an assay that could deliver results in less than 10 minutes per sample. A-TEEM spectroscopy captures both UV-Vis absorbance data and fluorescence excitation-emission information, producing a large spectral fingerprint for each sample. In this study, each sample generated more than 10,000 variables, which were then processed with multivariate models.

To build and test the system, the researchers split the data into calibration and test sets at a 75:25 ratio. That produced 1,816 samples for calibration and 584 for testing. They also took steps to avoid what they described as replicate trapping, keeping repeated measurements from the same original sample together so they would not artificially improve model performance during validation.

The best-performing model used support vector machine regression. The paper reported root mean square errors below 0.4 ng/kg for calibration, cross-validation and test prediction, with an R² of 0.879 for the test set and about 0.93 for calibration. The authors said those error levels were well below the lowest commonly cited sensory threshold for IBMP in red wine.

Using those results, the team estimated a detection limit of 1.32 ng/kg and a quantification limit of 4 ng/kg for the A-TEEM approach. That does not match the sensitivity of the most advanced GC-MS/MS methods using isotopically labeled standards, which can detect much lower concentrations, but it places the optical method in a range the researchers described as fit for purpose for screening grapes around relevant sensory thresholds.

The study also tested whether the method could sort samples above or below a practical decision point of 2 ng/kg. For that classification task, the researchers again used a support vector machine model. They reported a Matthews correlation coefficient of 0.970 and said there were no misclassified samples in the group below 2 ng/kg. The test set produced five false positives out of 584 samples, or less than 1%.

The authors noted that IBMP measured in grape extracts is expected to be somewhat higher than in finished wine, typically by around 30%, because the compound is readily extracted during winemaking. Even so, they argued that a threshold-based screening tool at the grape stage could still be useful for harvest decisions.

The paper places the new results against earlier work on A-TEEM prediction of IBMP in grape homogenates. In that earlier research, prediction errors and detection limits were higher than those reported here. The authors suggested several possible reasons for the improvement, including differences in sample sets and maturity conditions. In the previous study, grapes covered a wider range of ripeness levels and were sorted by berry density; in the new work, grapes were analyzed at harvest without sorting.

The researchers also said IBMP itself is unlikely to be directly responsible for fluorescence signals captured by A-TEEM. Instead, they believe the method detects a broader molecular fingerprint that correlates with IBMP concentration through other compounds present in grape extracts.

That point may matter for how the technology is used commercially. Because it is based on correlations rather than direct molecular identification of IBMP alone, it may be best suited as a rapid screening tool rather than a full replacement for confirmatory laboratory analysis. Still, its speed and lower expected cost could make it attractive for wineries handling large numbers of vineyard lots during harvest.

For the beverage sector, that could mean faster decisions on when to pick fruit and better control over unwanted vegetal character in wines made from susceptible varieties. In practice, a winery could use this kind of screening to flag lots near or above sensory risk levels before committing to harvest timing or blending plans.

The study was led by Adam Gilmore, Han Wang, David Jeffery, Ricardo Luna, Alvaro Gonzalez, Jorge Zincker and Monica Rodríguez Campos. It appears as a short communication in volume 60, issue 2 of OENO One. The paper was received on Feb. 9, accepted on April 30 and published on June 4.

The authors described the findings as evidence that A-TEEM combined with machine learning has potential as an inexpensive, rapid and accurate method for screening IBMP in extracts from red winegrape varieties that are vulnerable to green aroma character.